jotmi npd 2012

7/30/2019 jotmi npd 2012

1/16

20

ISSN: 0718-2724. (http://www.jotmi.org)Journal o Technology Management & Innovation Universidad Alberto Hurtado, Facultad de Economa y Negocios.

Received August 13, 2012 / Accepted December 04, 2012 J. Technol. Manag. Innov. 2012, Volume 7, Issue 4

Enhancing New Product Development (NPD) Portolio Perormance byShaping the Development Funnel

Paulo Soares Figueiredo1, Elisabeth Loiola2

Abstract

New product development (NPD) projects are typically managed through a series o screens, or gates, where ideas

compete or resources. Ideas are carved into projects, and these projects are reviewed, and approved or terminatedthrough the screening process so that only the best perorming projects continue to subsequent stages o design,

development and testing, and are released into the market place (Krishnan and Ulrich 2001; Terwiesch and Ulrich 2009).Most large innovative organizations deal with more than one NPD project at a time and typically engage in product pipeline

management (PPM), where a set o active projects are evaluated together while they traverse through a sequence o suchscreens. Key decisions in a R&D pipeline are: screen thresholds, complexity o projects, resource allocation and

capacity adjustment biases. We explore the impact o structural and behavioral aspects o these decisions through asimulation based analysis o a pharmaceutical dataset. Results establish concave relationships between value created atthe end o pipeline and the resource allocation and complexity allocation biases, indicating optimizability and a limit or

ront loading practices.

Keywords: product development, ideas, projects, product pipeline management, development unnel, stage/gate,screening.

1Senai Unidade Cimatec. Address: Av. Orlando Gomes, N 1845 - Piat Salvador-BA-Brazil Cep 41650-010 Tel: 71.3462-9500Fax: 71.3462-9599. Brazil. [email protected] Administration School, UFBA/NPGA. CNPq researcher. Brazil. [email protected]

7/30/2019 jotmi npd 2012

2/16

21


J. Technol. Manag. Innov. 2012, Volume 7, Issue 4

risk, and the resolution o uncertainty. Wheelwright andClark (1992) describe typical decision levers in this setting:

resource allocation (allocating workers), selection o taskcomplexity (dening number, size and relations betweentasks), capacity utilization (work intensity), and the level o

threshold (minimum quality or expected value) or passing

through a screen and the requency o screening. Resourceallocation dictates the types and amounts o resources avail-able or executing tasks beore a project, or a cohort o

projects, goes through a screen. Complexity selection de-nes the nature o these tasks, and the amount o resourcesit takes to complete these tasks. Even though the level o

complexity at each stage is predetermined to a certain de-gree by the existence o a minimum number o tasks to be

perormed, and their sequence, it is air to assume that thereis considerable reedom to managers while deciding project

activities. For example, Thomke and Fujimoto (2000) andKhurana and Rosenthal (1998) recommend the ront load-ing o activities in a project, i.e. the increase in complexity

and activities early in the development process, as a wayo reducing uncertainty and the amount o rework or new

work to be done later.

Capacity utilization aects the tradeo between outputquality (and thus the value created), and throughput. The to-

tal amount o resources available or allocation across stagesis determined by a budgeting exercise (Chao et al., 2009)and is divided among the stages, so that each stage receives

a raction o the total. The selection o average complexityin any one stage, on the other hand, is not subject to such a

global constraint. Product portolio management deals withthe problem o balancing these resources, complexity and al-

lied uncertainties, even without accounting or screening e-ects. Inclusion o screening eects, and balancing resourcesand project task complexity across stages o the pipeline

introduces time dependence into the portolio managementspace.

Accordingly, PPM is a multiaceted problem that refects the

dynamics o the product development pipeline, and the un-derstanding o how decisions variables interact to aect theshape o the pipeline (also called the unnel) and its innova-

tion outcomes. It is a normal practice within the portolioanalysis literature to convert the joint outcome o allocated

resources and selected complexity into the value created,which is typically measured in terms o net present value

(NPV) o new-product projects (Cooper et al, op. cit.). Thetrade-os between rate o value (or NPV) creation at eachstage that is dictated by the above mentioned decis ions and

throughput o the pipeline needs explicit ormulation andsystematic study. Thereore, the goal o this paper is to pro-

vide policy insights or PPM decis ion-making by analyzing theunderlying dynamics and determining which strategy allows

a higher perormance.

Introduction

New product development (NPD) projects are typicallymanaged through a series o screens, or gates, where ideascompete or resources. Ideas are carved into projects, and

these projects are reviewed, and approved or terminated

through the screening process so that only the best per-orming projects continue to subsequent stages o design,development and testing, and are released into the market

place (Krishnan and Ulrich 2001; Terwiesch and Ulrich 2009).Most large innovative organizations deal with more than oneNPD project at a time and typically engage in product pipe-

line management (PPM), where a set o active projects areevaluated together while they traverse through a sequence

o such screens.

Some empirical studies have explored the patterns, bestpractices or benchmarks in the managerial decisions con-cerning PPM (Schmidt et al., 2008). There is considerable

variation on the way such screening mechanisms are used.For instance, in the automotive industry, new platorms, such

as Chryslers mini-vans (e.g. the Plymouth Voyager in 1984)or Toyotas hybrid Prius (1997) are launched in an episodic

manner. Ater such launches, the market place expects thatnew models o these products will be launched on a peri-

odic, yearly basis, and it is rare in this industry to screenout an entire development project using the conventionalstage gate process described by Cooper et al. (1998). Other

industries may either oer more ungible product introduc-tion opportunities (e.g. sotware development managers re-

sort to versioning, see Cusumano and Selby, 1995), or maybe driven by long cycles o development (e.g. pharmaceu-

tical developments may have to screen out concepts sev-eral years ater the initiation o a discovery process basedon regulatory eedback (Hurtado, 2009) ). We steer away

rom analyzing conventional phase-gate processes that donot screen out products, and instead ocus on the unnels,

and especially their uzzy (uncertain) ront ends (Khuranaand Rosenthal, 1998; Zapata and Cant, 2008; Jugend and

Silva, 2012), where products have to go through undamen-tally dierent kinds o assessments across a succession oscreens that determinate the shape o the innovation unnel.

In our denition, the shape o the innovation unnel is estab-lished by the number o projects in play at various phases o

development awaiting a successive set o screens. Dierentindustries consider widely dierent numbers o innovations

at each phase or each innovation that is eventually launched(Terwiesch and Ulrich, op. cit., Figure 9-3).

How can managers improve their product pipeline peror-mance? Answers to this key strategic question will depend

on the ormulation o the process, associated decision struc-tures, and the denition o perormance, while explicitly re-

fecting elements o resource availability, task complexity,

7/30/2019 jotmi npd 2012

3/16

22



Theory

This section presents the theory used to develop the sys-

tem dynamics model o the product development pipelineused in the study. All the denitions presented here are keyto the understanding o the PPM model ormulation.

The process o product pipeline management has been or-mulated as a dynamic resource allocation problem that isoten beset with congestion eects due to resource con-

straints (Reinertsen 1997, Grin 1997, Cusumano and No-beoka 1998, Ulrich and Eppinger 2004). Decision rules orPPM have begun to be explored rom various viewpoints:

cycle time implications (Adler et al. 1995); stagewise re-source allocation (Banerjee and Hopp, 2001). More recently,

resource allocation insights have been ormulated rom abehavioral viewpoint, in terms o heuristics or resource

allocation across multiple stages o a pharmaceutical R&Dprocess (Gino and Pisano, 2005).

Allied studies within any one project have demonstrated theimportance o overlapping (Terwiesch and Loch 1999; Bhui-

yan et al., 2004; Marujo, 2009) and ront loading developmentactivities (Thomke and Fujimoto, 2000), meaning that more

time/resources are invested on each project on early stagesthus reducing uncertainty and subsequent rework. Thomke

and Fujimotos rationale is based on inormation exchangetheory that rewards early exchange o inormation and prob-lem solving. However, the ront-loading logic or a portolio

o projects diers rom its counterpart within a single pro-ject due to the aggregate considerations o resource alloca-

tion and complexity which are central to the PPM problem.The ront loading o average complexity means that tasks or

every project will be more complex and costly. This is a con-cept similar to the aorementioned one. The ront loading oglobal resources, however, is related to the overall capacity

to move projects to the second stage, and not to the com-plexity o a single project. In addition, at the portolio level

there is a need to consider the eects o screening thresh-olds and allied controls (e.g. resource utilization eciency)

that are also associated with these decisions.

Implications

A series o propositions or hypotheses to be tested in the

study is presented on the ollowing subsections, based onkey structural aspects o PPM and tradeos inherent to the

activity. However a ocus must be put on the key implica-tions rom the hypothesis testing process.

First and oremost, it is ound that both the allocation oresources and the selection o complexity exhibit convexity

that establishes optimal levels and limits to the advantageo ront loading across a portolio o products. This nding

is not obvious in the sense that in the available empirical

This PPM problem cannot be solved in closed orm (Ander-son and Morrice, 2006; Browning and Yassine, 2008). Fur-

thermore, the pipeline interactions are dicult to study withtraditional statistical methods or to anticipate with thoughtprocesses. A simulation study (Davis et al., 2007) is a par-

ticularly eective approach or tackling this problem because

a model o portolio value creation and throughput can beormulated to include multiple decision parameters and as-sociated longitudinal interactions within a process involving

multiple projects.

We developed a simulation model o the pipeline manage-

ment problem, and then calibrated and tested it with datarom a biotech company. The model is designed to deter-

mine guidelines or key simultaneous managerial decisions:choosing task complexity, allocating scarce resources and

adjusting product development capacity. It employs screenswith quality adjustment mechanisms, which select only thoseprojects with higher expected value (NPV). The model al-

lows the eort allocated to projects to be ront-loaded interms o average task complexity-- doing more work in an

early stage to characterize and justiy the projects. Becausesuch rontloading delays the projects proceeding to the

next stage o development, the shape o the pipeline, andthe backlogs rom one stage to the next, will consequently

change depending on how teams will adjust to the changesin workload. The model calculates the overall long-termpipeline perormance resulting rom dierent combinations

o decision parameters, in terms o the total expected NPVo completed projects. It also provides a mechanism to com-

pute the elasticity o the outcomes with respect to interme-diate thresholds and managerial choices.

The contributions rom our work are three old. First, weestablish a simulation-based methodology to assess pipeline

management and considerations o project throughput andquality. Second, we nd that both the allocation o resources

and the selection o complexity exhibit convexity that es-tablishes limits to the advantage o ront loading across a

portolio o products, i.e., there is an interior, optimal levelor these variables that is not on the extreme sides o theperormance curve. Third, the optimal level o resources and

project complexity depends on interactions o these vari-ables along with selected level o screens. We discuss the

managerial implications, both structural and behavioral, othese ndings.

We proceed as ollows: 2 presents theory on the pipelinemanagement phenomenon. 3 operationalizes the theory

and describes the PPM model. 4 identies the design o thesimulation study. 5 contains the statistical analysis o the

results rom the simulation study. Section 6 discusses mana-gerial implications o these results and limitations. Section 7

concludes.

7/30/2019 jotmi npd 2012

4/16

23



nominal utilization level, the increase in the value o the pro-jects as they pass through the gates should be proportionally

smaller, because the capacity utilization will be above or be-low its nominal levels (Girotra et al. 2005; Clark and Wheel-wright, 1993). This trade-o is represented on gure 1.

On the other hand, more projects can be released i theteams work more intensively. This suggests a competing hy-pothesis.

The capacity adjustment bias, a tendency that managersmight have while making the capacity adjustment decision,

aects value creation only in two cases: a) when there is apenalty or working more intensively, i.e. the capacity utiliza-

tion versus value creation curve is not fat (in such case itbecomes advantageous to produce more); and b) when the

pipeline has an overfow o projects and there are potentiallyhigh utilization rate(s), above 100%. When there is starvationo projects, the target capacity will always be the same (the

maximum possible), regardless o the direction o the bias.In other words, when the maximum capacity is below 100%,

it doesnt matter i managers have a bias towards achievingnominal capacity (100%) or towards reducing the backlogs

aster. The target capacities in both cases will be the same.Thereore it is proposed that as long as there is a potential

or high utilization rates (>100%) , an increase in the bias, atany one stage, towards reducing backlogs through the ad-

justment o capacity will 1) decrease the total value o the

projects due to lower value creation, and 2) increase valuedue to higher completion rates. I value creation is constant

(i.e. no tradeo or value creation), a bias towards reducingbacklogs will increase total value creation due to the pos-

sibility o enhancing the pipeline without penalty.

literature (e.g. Thomke and Fujimoto, 2000, Khurana andRosenthal, 1998), ront loading is recommended but there

is no tool to determine exactly how much or up to whichlevel it should be implemented. The existence o optimal lev-els is also not obvious since the problem cannot be solved

in closed orm, i.e. it is an NP hard problem (Anderson and

Morrice 2006; Browning and Yassine 2008).

Second, the optimal level o resources and project complex-

ity depends on interactions o these variables along withselected level o screens. In this study it was ound that thereare many signicant interactions in the process, which con-

tributed very signicantly to explain the variation on thedataset. Such interactions were not previously studied and

are signicant problem aced by managers. When the level oone variable is adjusted, the optimal level o other variables

may change. The simultaneous choice o these variables isurther complicated by such interactions.

We discuss next how these ndings were anticipated (andjustied) by both physical economical aspects o the PPM

process.

The Utilization Performance Trade Off

In a product development pipeline, the available capacity othe development teams is adjusted locally in order to eitheradapt to the work demand o each stage o the chain or

keep the utilization level around its nominal value. There-ore, we dene the capacity utilization bias as a managerial

tendency to work between these two extremes. The teamswill work more or less intensively depending on the capacity

adjustment choice. I more weight is given to the objective oachieving the astest rate to reduce backlogs instead o the

Figure 1: Relationship between Resource Utilization and NPV Creation Multiplier (Clark and Wheelwright, 1993)

*the nominal value o capacity utilization is set to be at unity.

7/30/2019 jotmi npd 2012

5/16

24



on the nature o the project portolio. In the sotware sec-tor, usually there are no such restrictions (McCormack et

al 2001). But or physical technologies in tangible products,certain project tasks must ollow others based on prece-dence relationship (Ford and Sterman, 1997), and all cannot

be perormed simultaneously upront. Such sequencing o

tasks reduces the average project complexity. Conversely,when other types o tasks can be pooled based on queu-ing considerations (Loch and Terwiesch, 1999) and executed

within a single stage, the average complexity rises. There arealso physical reasons or a limit to complexity allocation. Al-though a rm cannot always control the size o its projects,

the number o manhours per project is a variable that a-ects how intensively the teams will have to work, thereore

it aects capacity utilization. A longer time per project willresult in an increase o capacity utilization and/or decrease

in the number o projects released in the market. There-ore the relationship between project size and total valuecreated should also be concave. Thus, we propose the ol-

lowing hypothesis to complexity allocation: There is a limitto the perormance gains made by ront loading o average

complexity or a portolio o projects, while these projectstraverse across a product development pipeline.

Screening Threshold

The third set o decisions germane to the shape o the pipe-line is the level o competition between projects (Terwiesch

and Ulrich, 2009) and the associated screening threshold. Ar-guably, the higher the threshold, the lower will be the num-

ber o projects that pass through each o the screens and,accordingly, the lower would be the observed perormance

rom the resulting set o product innovations. Accordingly,we suggest that the perormance o the innovation pipelineis negatively associated with the screening threshold at each

stage.

However, the direct eect o screening threshold on theperormance does not complete the picture. Based on man-

agers allocation biases that are inormed by the level o thereward and allied degree o diculty (Frederick et al., 2002)there is endogeneity in the selection o screening levels and

the participants willingness to allocate resources and selectthe average task complexity. It is clear that depending on the

selectiveness o the screening process, the optimal alloca-tion o resources and complexity should change, because

thresholds aect the shape o the unnel and number o pro-jects at each stage. Thus, we make the ollowing proposition:The optimal level o resource allocation and complexity al-

location is mediated by the level o threshold at any onestage o the pipeline.

Resource Allocation

The market place rewards a steady delivery o projects,rather than a lumpy set o entries (Moorthy and PNG,1992). Thereore, it is reasonable to assume that a choice

o resource allocation ractions (basically how to allocate

people across stages) that balances the pipeline, ne tuningthe delivery o projects at its end will be preerable. Also,rom a quantitative viewpoint, the NPV o projects in the

late stages, being closer to the point o market launch, willexceed their early-stage counterpart. These considerationssuggest that there will be both physical and economic limits

to rontloading projects based on the need to construct atemporally balanced, high perormance portolio. The physi-

cal limits can be explained based on the assumption that therelationship between capacity utilization and value created

in each gate has an inverted U shape; it is intuitive to expectthat there exists an interior utilization level that maximizesrm prot. Utilization aects the dynamics across multiple

stages in a product development pipeline and managers othe NPD pipeline can reallocate resources across stages,

assigning people to other tasks in an attempt to optimizecapacity utilization globally. Hence, we oer the ollowing hy-

pothesis to the ront loading o resources: There is a limit tothe gains made by ront loading o resources or a portolio

o projects, while these projects traverse across a productdevelopment pipeline. We dene resource allocation bias astendency, while making the resource allocation decision, to

allocate more resources to one stage than to another.

Complexity Selection

While it is easy to see why resources should be allocatedto the ront end rom a problem-solving perspective or anyone project (increasing task complexity o the initial phase),

there is more to this picture (Cooper et al. 1998). All theactivities in a project cannot be concentrated in the early

stages. Indeed, behavioral evidence suggests that pipelinemanagers concentrate their attention on the last stage, due

to a combination o ocusing on recent events and alliedhero mentality (Repenning et al. 2001).

We dene average complexity as a measure o the num-ber o, magnitude o and relationship between tasks. This is

the metric that governs the nominal completion time, ora given amount o resources. Thereore, the average num-

ber o man-hours per project at each stage was used asa proxy or complexity in the model. Such proxy was alsoused by Yu, Figueiredo and Nascimento (2010). Indeed, allo-

cation o average complexity or a project, is a dual problem,with respect to resource allocation (Browning and Yassine

2008). However, the decision on how much complexity getsallocated to a stage o any one project (a managerial ten-

dency here called the complexity allocation bias) depends

7/30/2019 jotmi npd 2012

6/16

25



Model Description

The dynamics o PPM involve a broad decision context. Anymodel that tries to mimic or refect such decisions mustincorporate these key decision processes. The structure o

stocks and fows in PPM can be compared to the structure

o a service supply chain model (Anderson et al. 2005). Inboth situations, the processing fow-time and capacity con-straints have an impact on the throughput. The PPM problem

is thereore a special case o service supply chains, wheresome projects are terminated across stages based on theirNPV. Most rms use multiple gates in their pipelines (Ulrich

and Eppinger 2004). For parsimony, our model incorporatesonly the rst three gates, as shown in gure 2.

The basic structure and logic o the model are simple; every

month, a certain number o projects is started and entersthe pipeline. These projects are developed and screened insequence, beore being released into the marketplace. The

NPVs o the population o projects are tracked, enablingmanagers to decide how many projects will be terminated

and how much value will be lost due to termination. Val-ue creation happens while projects are developed at each

stage, and this value creation depends on how intensivelythe teams are working. Besides deciding on which projects

will be terminated (i.e., dening a screening threshold (T) orminimum allowable NPV or a project to be approved), man-agers also decide on three variables: the capacity adjustment

bias (), the resource allocation across stages (), and theaverage complexity o the projects (). Each o these vari-

The NPD Pipeline Model

This section begins with a description o the NPD pipelinemodel, which denes all the key terms to the understandingo the pipeline. A detailed description o the model, with a

list o equations can be ound in Figueiredo and Loiola (2012)

and Figueiredo and Joglekar (2007). The present study addstwo variables to the model which were treated as a constantin Figueiredo and Souza: The average complexity o projects

at each stage (measured by man-hours per project) and theraction o resources (Man-hours per month) that are al-located to each stage (see equations 1 and 2 below and the

original paper or more details). This section ends with anexplanation o the calibration and validation processes.

Variable (the resource allocation raction) is a value be-tween one and zero and is multiplied by the nominal capac-

ity at each stage. The sum o1, 2 and 3 is always equalto 100%. Thereore, resources can be allocated according do

dierent policies, refecting on the values o.

Figure 2: A Multi Stage Product Development Pipeline

7/30/2019 jotmi npd 2012

7/16

26



stages. The dynamic aspect o the pipeline adds complexityto the problem, and to the optimization eort.

Calibration

The model was calibrated to the Novartis Innovation Pipe-

line (Reyck et al. 2004). This case study has all the data nec-essary or the calibration, including NPV values at each stage,fows, complexity and resources. The Novartis pipeline has

our stages, but the rst stage (basic research) was excludedand only three stages were considered. The pipeline was cali-brated or a steady state condition, in which value creation is

maximum and there is a bias towards reducing the backlogs(=1). In the calibration procedure, the ollowing parame-

ters were kept constant, consistent with the data set: starts,resource allocation ractions, average Project Complexity

and Termination Rates. The ollowing output was matchedby perorming an iterative adjustment o the Gumbel untionlook up tables, by changing the mean gain and variance, while

keeping the nominal development times within reasonablerange: Average backlog in each stage and Average NPV in

each stage. The calibration achieved a goodness o t o 5%or all parameters, except the nominal development times.

Model Behavior and validation

Forrester and Senge (1980) suggest three types o tests osystem dynamics models: (1) structural similarity to the ac-

tual system; (2) reasonable behavior over a wide range o in-put values; and (3) behavior similarity to actual systems. The

ables aects capacity utilization (how intensively the teams

are working) and thereore the value creation, as describedon section 2.

The capacity adjustment bias refects managers tendenciesbetween either working aster/slower in order to reduce

the existing backlogs, or working at a constant rate (the bestor nominal rate) to optimize value creation (see gure 3).

The resource allocation bias refects managers tendenciesto allocate more people to work on the initial, mid or nalstages o the pipeline. Managers also have a bias towards

allocation o complexity, i.e. they can choose to increase/decrease the complexity o the projects in any stage o the

process. The perormance variables in the model are thetotal value created (NPV) at the end o the pipeline, value

creation rates at each stage and respective fows o projects.The adoption o NPV as the only perormance criteria orproject screening is a necessary simplication; in most com-

panies, however, more than one actor is used to enable the

decision to terminate a project, and dierent actors may beused depending on the stage o development o the project.For example, a pharmaceutical company might be more con-

cerned with the saety o a substance at the early stages andwith the manuacturability at later stages.The PPM problem is structured as a dynamic process in

the shape o a chain, thereore it is reasonable to assumethat accumulation and/or starvation might happen in such a

chain. Depending on the decisions made by managers, pro-jects may accumulate in early stages, or the later stages may

starve in case too many projects are terminated in early

Figure 3: Stock and Flow Structure o a Typical Stage

7/30/2019 jotmi npd 2012

8/16

27



Adding to these considerations, the models behavior isconsistent with the logics o product pipeline management

as ound in the literature. Also, by looking at gures 4, 5and 6, it is easy to see how the shape o the pipeline canbe determined not only by development time, but also by

complexity and resource allocation. For example, by ront

loading resources, the backlog o stage 1 is reduced, andthe backlog o stage 3 is increased, relative to the results othe back loading o resources. The opposite eect happens

when complexity is ront loaded. The eect o dierent con-gurations o the capacity adjustment bias is not signicantbecause in this study the pipeline is not overloaded with

projects, since the number o projects started is constantand balanced. When there was a larger infow o projects, it

was ound that a bias towards reducing the backlogs in actdoes reduce their sizes. Based on the above tests, the model

is considered useul or analysis o the product pipeline.

model was tested under extreme conditions (see gures 5to 7 and the next section) and it behaved well. Our experi-

mental design used a wide range o input values. Model be-havior remains reasonable through these tests. By basing themodel on the PPM literature and on service prot chains,

the models structural similarity to PPM as practiced is im-

proved. The data on actual product pipeline management al-lowed detailed model calibration and behavior comparison.The model behaved well in all studies.

In an attempt to increase external validity, a partial calibra-tion or a dataset rom Merck (Girotra et al. 2004) was re-

peated. In this calibration, only the average NPV values werenot known, so nominal values were used instead. Similar re-

sults were obtained; the most important dierence was thatall 3 stages had signicant parameters or Merck, instead o

only the rst stage or Novartis. The Merck calibration wasmodied by using the termination rates ound in Schmidtet al (2008) or an industrial setting, and obtained similar

results once again. It is clear that this study can be repeatedor many other settings, by making certain assumptions de-

pending on the availability o data.

Figure 4: Eect o Varying the Resource Allocation Bias

Figure 5: Eect o Varying the Complexity Selection Bias

Figure 6: Eect o Varying the Capacity Utilization Bias

Experimental Design

Linear regressions were used to analyze the simulation data-

set (as in Kleijnen 1995, Anderson et al 2005, Santos andSantos 2008). In this section, the specication o a regressionequation and the design o the numerical experiment are

presented.Recall that the model has been set up with the NPV

o the projects at the end o the third stage o the pipelineas the outcome variable that is shaped by resource alloca-tion, complexity selection, resource utilization raction, and

screening thresholds across the rst three stages. We usethis set up to speciy the ollowing regression equation:

7/30/2019 jotmi npd 2012

9/16

28



that higher thresholds or minimum NPV while screeningprojects will reduce the total value created. For the hypoth-

esis related to interaction between variables aecting per-ormance (hypothesis 4B): We expect C17, C18 and C19 tosignicant, which means that interactions are signicant and

the optimal level or one decision variable depends on the

value o other(s).

Model Testing

One experimental condition was conducted in order togenerate data. The condition was created to determine the

models behavior and generate policies in the presence o atradeo between capacity utilization and value creation.

The rst study or the experimental condition, a.k.a. our

base case, uses small perturbations to the parameters in thecalibrated model described in section 3. This study aims totest hypotheses 1 through 4. One sensitivity analysis study

is added to this base case: we run a scenario where there isno screening o projects, i.e. all projects are approved. This

study represents an extreme conditions test (Forrester andSenge, 1980), because potentially high utilization rates are

achieved, and ceiling eects might come into play in the pipe-line.

The design o the base case uses the nominal value rom thecalibrated model that was described in section 3 as deault

or medium parameters (Mk,K=2). See tables 1 and 2. The

nominal value o each decision parameter is then perturbed

to a higher and a lower setting in order to create variations,to create three values (M

k, K=1,3). For example, while al-

locating resources, the total nominal capacity o stages 1,2 and 3 remains xed. However, it can be divided unevenlyacross stages to perturb the variable That is, i1 is raised

by a actor o 5%, while keeping 2 constant, this means thatstage 3 will proportionally lower amount o resources. Then

a set o higher and lower values o1 is selected (Hk

andL

k, K=1,3) by adding or subtracting 5%. These parameters

ensure that each segment o the utilization curve is sampled,without hitting extreme conditions. The selected complexity,i.e. needed number o man-hours per project in each stage is

represented by decision variable Gamma (i). The high andlow conditions are calculated by adding or subtracting 5%

to i, 10% to i and i and 0.5 standard deviations o theNPVs to the thresholds (Ti). These amounts o perturba-

tions have been selected to span a large range o settings orthe decision variables, while satisying the constraints im-posed by the table unction used to calculate NPV creation

rate based on utilization. For example, an increase o 5% in1 results in an increase o 32% in the capacity o stage I.

This yields a partial actorial 7x7x7x7 design (see table 3),

that requires a total o 2401 (i.e. NBase

= 2401) simulations.

Herei =1 2 3 represents stages o the pipeline

Cj = Regression coecients, j=1, j=2j=19V: Total NPV at the end o the third stage o thepipeline at the end o planning horizon

: Noise Terms

Decisions

i : increase in this parameter enhances resource allocationraction or stages i, such that (i >= 0 and 1+ 2+ 3=1)

i : Complexity at stage i, such that i >= 0i: Capacity adjustment bias at stage i (0

7/30/2019 jotmi npd 2012

10/16

29



Variable Stage 1 Stage 2 Stage 3

NPV gain rate 3.3839 1.2599 3.1128

Average NPV (MUS$) 82.07 277.74 349.93

Backlog (projects) 37 60 5

Resource Allocation fraction () 0.1579 0.3816 0.4605

Complexity () manhours/project* 112000 1082667 5226667

Starts 40 projects/year 10%

Termination Rates 75% 75% 20%

Table 1: Calibrated Parameters in the Base Case

*Assumed cost o manhour o $75

Table 2: Flow Parameters the Base Case

Table 3: Experimental Design - Base case (n=2401)

*M=base case values, L=Low values, H=High values

Stage 1 2 3Flow: Base Case 10.1 2.59 2.146

, ,T Stage 1 Stage 2 Stage 31 M1 M2 M32 H1 M2 M33 L1 M2 M34 M1 H2 M35 M1 L2 M36 M1 M2 H37 M1 M2 L3

Similar designs are repeated or the other studies. We notethat the sensitivity study does not admit variation in thresh-

olds. That is, NNo_Screens =343. The correlation matrix orthe regression variables in the base case, is shown in Table 4.

Results

Experimental Condition 1: Capacity Utilization vs

Value Creation Tradeoff

For the base case o condition 1, the regressions that ex-amine hypotheses 1 through 4 are reported in Table 5. The

sensitivity analysis results are reported in Table 6. Models 1through 5 (corresponding to columns 1 to 5 respectively)

examine perturbations to the base case. We only report thetwo most signicant interactions.

We now review the results shown in models 3-5. Hypoth-esis 1 deals with diminishing returns on the variable i. Co-

ecient C1 is positive and signicant and C2 is negative andsignicant. . On the other hand, resource allocation or stage

2 is not signicant. This suggests that or the set o base caseparameters, the queuing physics o the pipeline makes the

tuning and optimization o the rst stage more important

than that o the other stages. Thus, hypothesis 1 is supportedor stage 1 in the base case.

Hypothesis 2 deals with diminishing returns on the variablei. Coecient C5 is positive and signicant and C6 is nega-

tive and signicant. Thus, this hypothesis is supported orstage 1 in the base case. Only the coecients or the rst

stage are signicant.

Hypothesis 3 deals with the utilization perormance tradeo. This hypothesis is not supported in the base case, be-cause coecients C11, C12 and C13 are not signicant.

However, coecients C12 and C13 were signicant on thesensitivity analysis study. The deviation o results rom the

hypothesized pattern o behavior is urther discussed in 5.2.Hypothesis 4A and 4B are related to the eect o the

screening thresholds. Hypothesis 4A is supported becausecoecients C14, C15 and C16 are signicant and negativeas expected. That is, higher thresholds result in more ter-

mination and value loss, as expected. Hypothesis 4B is par-tially conrmed: the optimal level o resource allocation is

mediated by the level o threshold at the rst stage o thepipeline. The level o complexity, on the other hand, is not

mediated by the threshold since this interaction was not

7/30/2019 jotmi npd 2012

11/16

30



T1 T2 T3Al-

pha1

Al-

pha2

Al-

pha3

Gam-

ma1

Gam-

ma2

Gam-

ma3Beta1 Beta2 Beta1sq Beta2sq

Gam-

ma1sq

Gam-

ma2sq

Gam-

ma3sq

T1 1

T2 .000 1

T3 .000 . 000 1

Alpha1 .000 .000 .000 1

Alpha2 .000 .000 .000 .000 1

Alpha3 .000 .000 .000 .000 .000 1Gamma1 .000 .000 .000 .000 .000 .000 1

Gamma2 .000 .000 .000 .000 .000 .000 .000 1

Gamma3 .000 .000 .000 .000 .000 .000 .000 .000 1

Beta1 .000 .000 .000 .000 .000 .000 .000 .000 .000 1

Beta2 .000 .000 .000 .000 .000 .000 .000 .000 .000 .200** 1

Beta1sq .000 .000 .000 .000 .000 .000 .000 .000 .000 .994** .199** 1

Beta2sq .000 .000 .000 .000 .000 .000 .000 .000 .000 .200** .999** .195** 1

Gamma1sq .000 .000 .000 .000 .000 .000 .999** .000 .000 .000 .000 .000 .000 1

Gamma2sq .000 .000 .000 .000 .000 .000 .000 .999** .000 .000 .000 .000 .000 -.001 1

Gamma3sq .000 .000 .000 .000 .000 .000 .000 .000 .999** .000 .000 .000 .000 -.001 -.001 1

Table 4: Correlations Matrix Base Case

Variable Treatment 1 2 3 4 56 W/O

Screening

Constant

1(Cap. Adjustment Bias)

22712.26*

(60.39)

31405.133*

(60.39)

1 (Front Loaded Resource)n/a

-1864.89

70085.237* 70085.237* 103409.699* 27639.865* 50115.406*

-1860.85 -6541.69 -3392.77 (2495.15)

1 Squaredn/a

-5882.17

-212751.85* -212751.85* -212751.85* -212751.85* -150219.294*

-5869.45 -5836.24 -5621.27 (7870.12)

2 (Mid Loaded Resource) n/a

2 Squared n/a

1 (Front Loaded Complexity) n/an/a

-0.028

0.079* 0.079*

-0.028

1 Squared n/a

n/a

0

-3.5E-7* -3.501E-7* -3.501E-7*

0 0

2 (Mid Loaded Complex.) n/a n/a

2 Squared n/a n/a

3(Back Loaded Complex.) n/a n/a

3 Squared n/a n/a

T1

-85.52

-566.988* -566.988* -566.988* 1108.786* -566.988* N/A

-60.18 -60.05 -321.13 -57.51

7/30/2019 jotmi npd 2012

12/16

31



Table 5: Results Base Case, Condition 2* P

7/30/2019 jotmi npd 2012

13/16

32



Screening and Starting, Value Creation and Screen-

ing Policies Shift Bottlenecks

Results rom models 3-5 and 6, show that the PPM prob-lem is not a conventional queuing problem, in the sense that

managers are more interested in the value o the through-

put, rather than just the waiting time. Moreover, there aretwo key state variables that co-fow (Sterman 2000): backlogand the total value o the portolio at any one stage. In this

structure, either the absence or the level o screens can shitthe completion rates as shown in Table 7. Managerially, theimplication or such shit implies that there is need to track

the product resource allocation, complexity and NPV crea-tion rates by stages. These data can be use to balance the

pipeline and to change the screening levels endogenously ithe pipeline is not balanced.

Another important eect o dynamics deals with the localbiases captured in terms o the local capacity adjustment

parameter (i). In the case where there was no screening,potentially high capacity utilization are reached in stages 2

and 3, and the policy o working more intensively to reducethe backlogs is the best choice.

Conclusion

This study develops a set o hypotheses on the shaping anddynamics o the product development pipeline. A model is

developed to assess pipeline management considerations oproject throughput and quality, and data by means o simula-

tion. The model is calibrated with data rom a pharmaceuti-cal setting. We nd that the allocation o resources and the

task complexity exhibit convexity that establishes optimiz-ability and limits to the advantage o ront loading acrossa portolio o products. The optimal level o resources and

project complexity depends on interactions o these varia-bles along with selected level o screens and other variables.

Ours is a highly stylized model that comes with several limi-

tations. For instance, we do not account or dependenciesamong projects, such as sharing o resources and sub-ad-ditive pay-os. We have shown the model set up to port-

olio managers in the bio-tech industry. They have pointedout that the model depends crucially upon the availability

o reliable data on NPV estimation, and allied value creationrates, especially early in the pipeline. A second limitation is

the manner in which some rms handle the idea generationprocess and dene projects. Complexity in such setting maynot be selected in an independent manner certain projects

must be lumped together while others must ollow set se-quences, and yet other set o tasks are viewed as platorms

or may involve rework. Our model is aggregate and does notaccount or these eects. Moreover, shared resources aect

the relationship between development costs and project de-

Rosenthal 1997) and recommends a ront loading strategy(Thomke and Fujimoto 2000) or a development process.

These NPD studies have been empirical in their orientationand based on these observations, their authors argue orthe need to ocus on the ront end owing either to the high

amount o uncertainty or to the ability to generate early

inormation. Consistent with these NPD ndings (which re-er to single projects), but based on pipeline managementand throughput-quality tradeo considerations, our analy-

sis shows that management o ront end, i.e. stage 1, isimportant in governing the overall pipeline perormance.Contrary to existing practices on multiproduct management

(Repenning et al. 2001), the management o the uzzy rontend matters more to long term perormance than trying to

ocus on the late stages o the pipeline.

Accounting for Convexity and Interactions during

Pipeline Design

Conrmations o hypotheses 1 and 2 indicate that rontloading o resources and average complexity selection in

stage 1 are jointly concave in respect to pipeline peror-mance. This means that these variables are optimizable and

there are diminishing returns or them, as explained in sec-tion 2. Moreover, the results are dependent on the choice o

thresholds and their interaction with the resources in stage1 (as indicated by the conrmation o hypotheses 4). A man-agerial implication o these results is: when endowed with all

the data, managers can ne tune the perormance o theirrespective pipeline by ollowing our procedure. Managers

should be cognizant o the interactions between resources,complexity and thresholds, particularly at the ront end o

the pipeline. Such interactions are not accounted or in con-ventional portolio analysis (Terwiesch and Ulrich 2009). Thesignicance o interactions show that it is important to ne

tune the pipeline in a consistent, holistic manner, because achange in one o the variables must be ollowed by a change

in others, to account or the modication in the utilizationrate and consequently in the value created. I one variable is

adjusted, it can change the optimal level o others.

These ndings also ampliy the need or research on ormal

models or the pipeline outcome at any one stage, in termso resources, complexity o screening threshold, particularly

when that stage is the bottleneck. Similarly, the behavioralimplications o the ndings in terms o the propensity o

an individual team at any one stage to pick higher or lowercomplexity, when given a xed amount o resource, and giv-en the historical probability o getting through a particular

screen. We leave both these aspects as items or ollow onwork.

7/30/2019 jotmi npd 2012

14/16

33



CUSUMANO, M. & Nobeoka K. (1998). Thinking BeyondLean, Freepress.

DAHAN, E., and Mendelson, H. (2001). An extreme-valuemodel o concept testing. Management Science, 47(1), 102-

116.

DAVIS, J.P., Eisenhardt, K.M. and Bingham, C.B. 2007. Devel-

oping theory with simulation methods. Academy o Manage-ment Review, 32(2), 580-599.

DEEDS, D. L., & Rothaermel, F. T. (2003). Honeymoons andliabilities: The relationship between age and perormance in

research and development alliances. Journal o Product In-novation Management, 20(6), 468-484.

FIGUEIREDO, P., E. Loiola (2012): A Cofow structure toScreen Projects in a Product Pipeline. UFBA.BR Working pa-

per. Available rom the author.

FIGUEIREDO,P. Joglekar, N. (2007): Dynamics o ProjectScreening in a Product Development Pipeline. In: 25th In-

ternational Conerence o The System Dynamics Society,Boston.

http://www.systemdynamics.org/conerences/2007/pro-ceed/papers/JOGLE201.pd [accessed July 2012].

FORD, D. & Sterman, J. D. (1997) Dynamic modeling o prod-uct development processes. System Dynamics Review, 14,

31-68.

FORRESTER, J. W. and Senge, P. M. (1980). Tests or buildingcondence in system dynamics models. TIME Studies in theManagement Science 14, 209-228.

FREDERICK, S., Loewenstein, G. & ODonoghue, T. (2002).

Time discounting and time preerence: A critical review.Journal o Economic Literature. 40, 351-401.

GALAMBOS, Janos (1978). The asymptotic theory o Ex-treme Order Statistics. John Wyley and Sons.

GINO, F., Pisano, G. (2005). Do Managers Heuristics Aect

R&D Perormance Volatility? A Simulation Inormed by thePharmaceutical Industry. Harvard Business School Working

Paper.

GIROTRA, K., C. Terwiesch, K. T. Ulrich (2004). New Drug

Development at Merck & Co., Case Study, The WhartonSchool.

GIROTRA, K., Terwisch, C., Ulrich, K.T. (2005). Managing the

Risk o Development Failures: A Study o Late-Stage ailures

velopment time (Girotra et al. 2005). Sub-additive pay-osoccur when a rm launches many products that are related

(such as derivatives o a product amily), but could generatethe same revenues i it had developed only one product.Developing ormal models o the economics o screening, in

the presence o complexity and resource tradeos, either at

a single stage or in a cascade o stages, and accounting orbehavior bias (Gino and Pisano, 2005) oers opportunitiesor ollow on work.

References

ADLER, P. S., Mandelbaum, A., Nguyen, V., & Schwerer, E.(1995). From project to process management - An empir-

ically-based ramework or analyzing product developmenttime. Management Science, 41(3), 458-484.

ANDERSON, E. G., Morrice, D. J., & Lundeen, G. (2005). Thephysics o capacity and backlog management in service

and custom manuacturing supply chains. System DynamicsReview, 21(3), 217-247.

ANDERSON, E. G., Morrice, D.J (2006). Stochastic Opti-

mal Control o Centralized Stang and Backlog Policies in aTwo-Stage Customized Service Supply Chain. Production

and Operations Management, 15 (2): 263-278.

BHUIYAN, N., Gerwin, D., Thomson, V., (2004) Simulation o

the New Product Development Process or PerormanceImprovement. Management Science, 50(12): 1690-1703.

BROWNING T., Yassine, A. (2008). Analyzing Multiple Prod-

uct Development Projects Based On Inormation and Re-source Constraints. MIT Working paper.http://libaxis5.mit.edu/bitstream/handle/1721.1/1670/multi-

DSM.pd?sequence=1

BANERJEE, S., Hopp, W. J. (2001). The Project Portolio Man-agement Problem. Department o Industrial Engineering

and Management Sciences, Northwestern University, June.

CHAO, R. O., Kavadias, S., Gaimon, C. (2009) Revenue Driv-

en Resource Allocation: The Eects o Organization Designand Incentives on NPD Portolio Management. Management

Science. 55(9): 1556-1569.

CLARK, K. B., Wheelwright, S. C. (1993). Managing NewProduct and Process Development. Free Press, NY.

COOPER, R.G., Edgett S.J., Kleinschmidt E.J. (1998): PortolioManagement or New Products. 2nd Ed. Perseus Publishing,

MA.

CUSUMANO, M. J. and Selby, R. W. (1995) Microsot Secrets,HarperCollins.

7/30/2019 jotmi npd 2012

15/16

34



and Congestion Eects, Journal o Product Innovation Man-agement, 16(2): 145-159.

MARUJO, L.G. (2009). Rework Impacts Evaluation throughSystem Dynamics Approach in

Overlapped Product Development Schedule, J. Technol.

Manag. Innov. Volume 4, Issue 2.

MCCORMACK, A., Verganti R., and M. Iansiti (2001), De-

veloping Products on Internet Time: The Anatomy o aFlexible Development Process, Management Science, 47(1):133-150.

MOORTHY, K. S., I. P. L. Png.(1992) Market segmentation,

cannibalization, and the timing o product introductions,Management Science, 38(3): 345-359.

REINERTSEN D.G. (1997). Managing the Design Factory. Si-mon and Schuster, New York, New York.

REPENNING, N. (2001). Understanding Fire Fighting in New

Product Development, Journal o Product Innovation Man-agement, 18(5): 285-300.

REYCK, B., Degraeve, Z. and Crama, P. (2004): Project Port-

olio Management at Novartis Pharma. London BusinessSchool case.

SANTOS, P.M.R. and Santos, M.I.R (2008). Using subsystemlinear regression metamodels in stochastic s imulation. Euro-

pean Journal o Operational Research. Vol. 5, N. 5.

SCHMIDT, J.B., Sarangee, K., Montoya-weiss, M.M. (2008).Exploring New Product Development Project Review Prac-tices and Perormance. Drat, being revised at Journal o

Product Innovation Management.

STERMAN, J. (2000) Business Dynamics: Systems Think-ing and Modeling or a Complex World. New York: Irwin/

McGraw-Hill.

TERWIESCH, Christian, Christoph H. Loch (1999). Meas-

uring the Eectiveness o Overlapping Development Activi-ties, Management Science, 45(4): 455-465;

TERWIESCH, Christian, Karl T. Ulrich (2009). Innovation

Tournaments: Creating and Selecting Exceptional Opportu-nities, Harvard Business School Press.

THOMKE, S., & Fujimoto, T. (2000). The eect o ront-load-ing problem-solving on product development perormance.

Journal o Product Innovation Management, 17(2): 128-142.

in the Pharmaceutical Industry. The Wharton School, Univer-sity o Pennsylvania, January.

GIROTRA, K., Terwiesch, C. and Ulrich, K., Idea Generationand the Quality o the Best Idea (June 15, 2009). INSEAD

Business School Research Paper No. 2009/32/TOM. Avail-

able at SSRN: http://ssrn.com/abstract=1082392

GREENE, W.H. (1997). Econometric Analysis, 3rd Edition.

Upper Saddle River, NJ: Prentice-Hall.

GRIFFIN, A. (1997). PDMA research on new product devel-

opment practices: Updating trends and benchmarking bestpractices. Journal o Product Innovation Management, 14(6),

429-458.

GUMBEL, E. J. (1958): Statistics o Extremes. Columbia Uni-versity Press, New York.

KAVADIAS S. and Loch, C.H. (2004). Project Selection UnderUncertainty: Dynamically Allocating Resources to Maximize

Value. Hillier International Series in Operations Researchand Management Science, Kluwer Academic Publishers (now

Springer-Verlag), Boston, MA.

HESKETT, J.L., Sasser, W.E., Schlesinger, L.A. (1997). The Ser-vice Prot Chain. New York: Free Press.

HURTADO, C.N. (2009): Biogenricos: Un Estudio de Vigi-lancia Tecnolgica para el Caso de La Situacin en Chile J.

Technol. Manag. Innov., Volume 4, Issue 3.

JUGEND, D. and Silva, S.L. (2012): Integration in New Prod-uct Development: Case Study in a Large Brazilian High-Tech-nology Company. J. Technol. Manag Innov., Volume 7, Issue 1.

KHURANA, A., & Rosenthal, S. R. (1997). Integrating the

uzzy ront end o new product development. Sloan Manage-ment Review, 38(2), 103-120.

KLEIJNEN (1995). Sensitivity Analysis and optimization oSystem Dynamics models: Regression Analysis and Statistical

Design o Experiments. System Dynamics Review, Vol 11,n.4, 275-288.

KRISHNAN V., Ulrich, K. (2001). Product development deci-

sions: a review o the literature. Management Science. 47 i1.1-21.

LAW, A.M., Kelton, W.D. (2000). Simulation Modeling andAnalysis (3rd ed.) McGraw-Hill: New York.

LOCH, Christoph H., Christian Terwiesch, (1999). Acceler-

ating the Process o Engineering Change Orders: Capacity

7/30/2019 jotmi npd 2012

16/16



ULRICH,K.T., Eppinger, S.D. (2004). Product Design and De-velopment. Third Edition, McGraw-Hill, New York.

WHEELWRIGHT, S.C. and Clark, K. B. (1992). Revolution-izing Product Development: Quantum Leaps in Speed, E-

ciency and Quality. The Free Press.

YU, A.S.O., Figueiredo, P.S, de Souza Nascimento, P. T. (2010):Development Resource Planning: Complexity o Product

Development and the Capacity to Launch New Products.Journal o Product Innovation Management. 27(2): 253-266

ZAPATA, A.R.P, Cant, S.O. (2008): Gestion Estrategica DeLa Tecnologia En El Predesarrollo De Nuevos Productos. J.

Technol. Manag. Innov. Volume 3, Issue 3.

Appendix

Linear term o the quadratic equations.

I perormance is a quadratic equation o a variable x (com-plexity or resources) :

Value Created = Ax2+Bx+C (where A is negative and x ispositive).

In the optimal point, where perormance is maximum, wehave:

0=2Ax+B so B=-2Ax. But we know that A0 at thispoint, thereore B>0.

Documents

jotmi npd 2012